1 /*
2  * Copyright (C) 2011 Red Hat, Inc.
3  *
4  * This file is released under the GPL.
5  */
6 
7 #include "dm-btree-internal.h"
8 #include "dm-transaction-manager.h"
9 
10 #include <linux/device-mapper.h>
11 
12 #define DM_MSG_PREFIX "btree spine"
13 
14 /*----------------------------------------------------------------*/
15 
16 #define BTREE_CSUM_XOR 121107
17 
18 static int node_check(struct dm_block_validator *v,
19 		      struct dm_block *b,
20 		      size_t block_size);
21 
node_prepare_for_write(struct dm_block_validator * v,struct dm_block * b,size_t block_size)22 static void node_prepare_for_write(struct dm_block_validator *v,
23 				   struct dm_block *b,
24 				   size_t block_size)
25 {
26 	struct btree_node *n = dm_block_data(b);
27 	struct node_header *h = &n->header;
28 
29 	h->blocknr = cpu_to_le64(dm_block_location(b));
30 	h->csum = cpu_to_le32(dm_bm_checksum(&h->flags,
31 					     block_size - sizeof(__le32),
32 					     BTREE_CSUM_XOR));
33 
34 	BUG_ON(node_check(v, b, 4096));
35 }
36 
node_check(struct dm_block_validator * v,struct dm_block * b,size_t block_size)37 static int node_check(struct dm_block_validator *v,
38 		      struct dm_block *b,
39 		      size_t block_size)
40 {
41 	struct btree_node *n = dm_block_data(b);
42 	struct node_header *h = &n->header;
43 	size_t value_size;
44 	__le32 csum_disk;
45 	uint32_t flags;
46 
47 	if (dm_block_location(b) != le64_to_cpu(h->blocknr)) {
48 		DMERR_LIMIT("node_check failed: blocknr %llu != wanted %llu",
49 			    le64_to_cpu(h->blocknr), dm_block_location(b));
50 		return -ENOTBLK;
51 	}
52 
53 	csum_disk = cpu_to_le32(dm_bm_checksum(&h->flags,
54 					       block_size - sizeof(__le32),
55 					       BTREE_CSUM_XOR));
56 	if (csum_disk != h->csum) {
57 		DMERR_LIMIT("node_check failed: csum %u != wanted %u",
58 			    le32_to_cpu(csum_disk), le32_to_cpu(h->csum));
59 		return -EILSEQ;
60 	}
61 
62 	value_size = le32_to_cpu(h->value_size);
63 
64 	if (sizeof(struct node_header) +
65 	    (sizeof(__le64) + value_size) * le32_to_cpu(h->max_entries) > block_size) {
66 		DMERR_LIMIT("node_check failed: max_entries too large");
67 		return -EILSEQ;
68 	}
69 
70 	if (le32_to_cpu(h->nr_entries) > le32_to_cpu(h->max_entries)) {
71 		DMERR_LIMIT("node_check failed: too many entries");
72 		return -EILSEQ;
73 	}
74 
75 	/*
76 	 * The node must be either INTERNAL or LEAF.
77 	 */
78 	flags = le32_to_cpu(h->flags);
79 	if (!(flags & INTERNAL_NODE) && !(flags & LEAF_NODE)) {
80 		DMERR_LIMIT("node_check failed: node is neither INTERNAL or LEAF");
81 		return -EILSEQ;
82 	}
83 
84 	return 0;
85 }
86 
87 struct dm_block_validator btree_node_validator = {
88 	.name = "btree_node",
89 	.prepare_for_write = node_prepare_for_write,
90 	.check = node_check
91 };
92 
93 /*----------------------------------------------------------------*/
94 
bn_read_lock(struct dm_btree_info * info,dm_block_t b,struct dm_block ** result)95 int bn_read_lock(struct dm_btree_info *info, dm_block_t b,
96 		 struct dm_block **result)
97 {
98 	return dm_tm_read_lock(info->tm, b, &btree_node_validator, result);
99 }
100 
bn_shadow(struct dm_btree_info * info,dm_block_t orig,struct dm_btree_value_type * vt,struct dm_block ** result)101 static int bn_shadow(struct dm_btree_info *info, dm_block_t orig,
102 	      struct dm_btree_value_type *vt,
103 	      struct dm_block **result)
104 {
105 	int r, inc;
106 
107 	r = dm_tm_shadow_block(info->tm, orig, &btree_node_validator,
108 			       result, &inc);
109 	if (!r && inc)
110 		inc_children(info->tm, dm_block_data(*result), vt);
111 
112 	return r;
113 }
114 
new_block(struct dm_btree_info * info,struct dm_block ** result)115 int new_block(struct dm_btree_info *info, struct dm_block **result)
116 {
117 	return dm_tm_new_block(info->tm, &btree_node_validator, result);
118 }
119 
unlock_block(struct dm_btree_info * info,struct dm_block * b)120 void unlock_block(struct dm_btree_info *info, struct dm_block *b)
121 {
122 	dm_tm_unlock(info->tm, b);
123 }
124 
125 /*----------------------------------------------------------------*/
126 
init_ro_spine(struct ro_spine * s,struct dm_btree_info * info)127 void init_ro_spine(struct ro_spine *s, struct dm_btree_info *info)
128 {
129 	s->info = info;
130 	s->count = 0;
131 	s->nodes[0] = NULL;
132 	s->nodes[1] = NULL;
133 }
134 
exit_ro_spine(struct ro_spine * s)135 int exit_ro_spine(struct ro_spine *s)
136 {
137 	int r = 0, i;
138 
139 	for (i = 0; i < s->count; i++) {
140 		unlock_block(s->info, s->nodes[i]);
141 	}
142 
143 	return r;
144 }
145 
ro_step(struct ro_spine * s,dm_block_t new_child)146 int ro_step(struct ro_spine *s, dm_block_t new_child)
147 {
148 	int r;
149 
150 	if (s->count == 2) {
151 		unlock_block(s->info, s->nodes[0]);
152 		s->nodes[0] = s->nodes[1];
153 		s->count--;
154 	}
155 
156 	r = bn_read_lock(s->info, new_child, s->nodes + s->count);
157 	if (!r)
158 		s->count++;
159 
160 	return r;
161 }
162 
ro_pop(struct ro_spine * s)163 void ro_pop(struct ro_spine *s)
164 {
165 	BUG_ON(!s->count);
166 	--s->count;
167 	unlock_block(s->info, s->nodes[s->count]);
168 }
169 
ro_node(struct ro_spine * s)170 struct btree_node *ro_node(struct ro_spine *s)
171 {
172 	struct dm_block *block;
173 
174 	BUG_ON(!s->count);
175 	block = s->nodes[s->count - 1];
176 
177 	return dm_block_data(block);
178 }
179 
180 /*----------------------------------------------------------------*/
181 
init_shadow_spine(struct shadow_spine * s,struct dm_btree_info * info)182 void init_shadow_spine(struct shadow_spine *s, struct dm_btree_info *info)
183 {
184 	s->info = info;
185 	s->count = 0;
186 }
187 
exit_shadow_spine(struct shadow_spine * s)188 int exit_shadow_spine(struct shadow_spine *s)
189 {
190 	int r = 0, i;
191 
192 	for (i = 0; i < s->count; i++) {
193 		unlock_block(s->info, s->nodes[i]);
194 	}
195 
196 	return r;
197 }
198 
shadow_step(struct shadow_spine * s,dm_block_t b,struct dm_btree_value_type * vt)199 int shadow_step(struct shadow_spine *s, dm_block_t b,
200 		struct dm_btree_value_type *vt)
201 {
202 	int r;
203 
204 	if (s->count == 2) {
205 		unlock_block(s->info, s->nodes[0]);
206 		s->nodes[0] = s->nodes[1];
207 		s->count--;
208 	}
209 
210 	r = bn_shadow(s->info, b, vt, s->nodes + s->count);
211 	if (!r) {
212 		if (!s->count)
213 			s->root = dm_block_location(s->nodes[0]);
214 
215 		s->count++;
216 	}
217 
218 	return r;
219 }
220 
shadow_current(struct shadow_spine * s)221 struct dm_block *shadow_current(struct shadow_spine *s)
222 {
223 	BUG_ON(!s->count);
224 
225 	return s->nodes[s->count - 1];
226 }
227 
shadow_parent(struct shadow_spine * s)228 struct dm_block *shadow_parent(struct shadow_spine *s)
229 {
230 	BUG_ON(s->count != 2);
231 
232 	return s->count == 2 ? s->nodes[0] : NULL;
233 }
234 
shadow_has_parent(struct shadow_spine * s)235 int shadow_has_parent(struct shadow_spine *s)
236 {
237 	return s->count >= 2;
238 }
239 
shadow_root(struct shadow_spine * s)240 int shadow_root(struct shadow_spine *s)
241 {
242 	return s->root;
243 }
244 
le64_inc(void * context,const void * value_le)245 static void le64_inc(void *context, const void *value_le)
246 {
247 	struct dm_transaction_manager *tm = context;
248 	__le64 v_le;
249 
250 	memcpy(&v_le, value_le, sizeof(v_le));
251 	dm_tm_inc(tm, le64_to_cpu(v_le));
252 }
253 
le64_dec(void * context,const void * value_le)254 static void le64_dec(void *context, const void *value_le)
255 {
256 	struct dm_transaction_manager *tm = context;
257 	__le64 v_le;
258 
259 	memcpy(&v_le, value_le, sizeof(v_le));
260 	dm_tm_dec(tm, le64_to_cpu(v_le));
261 }
262 
le64_equal(void * context,const void * value1_le,const void * value2_le)263 static int le64_equal(void *context, const void *value1_le, const void *value2_le)
264 {
265 	__le64 v1_le, v2_le;
266 
267 	memcpy(&v1_le, value1_le, sizeof(v1_le));
268 	memcpy(&v2_le, value2_le, sizeof(v2_le));
269 	return v1_le == v2_le;
270 }
271 
init_le64_type(struct dm_transaction_manager * tm,struct dm_btree_value_type * vt)272 void init_le64_type(struct dm_transaction_manager *tm,
273 		    struct dm_btree_value_type *vt)
274 {
275 	vt->context = tm;
276 	vt->size = sizeof(__le64);
277 	vt->inc = le64_inc;
278 	vt->dec = le64_dec;
279 	vt->equal = le64_equal;
280 }
281